openmp相关，请教一下问题出在哪里导致程序效率无法提升？

[zhongxyui]

int SA(float*HFt,float*HFb,float*HMt,float*HMb,float*Vop,float*Vos,float*Vpmax,float*Vpmin,float*Vsmax,float*Vsmin,float*Xr,float*Yr,float*Hr,float*top,float*tos,int numlayer,int numreceiver,float Hsource,float Xsource,float Ysource,float leightmodel,int numbest,float weight)
{
	int i,j,k;
	int markovlength=10000;
	float decayscale=0.95;
	float stepfactor=0.2;
	float temperature=100;
	float tolerance=0.1;
	float*preVp=(float*)malloc(numlayer*sizeof(float));
	float*nextVp=(float*)malloc(numlayer*sizeof(float));
	float*prebestVp=(float*)malloc(numlayer*sizeof(float));
	float*bestVp=(float*)malloc(numlayer*sizeof(float));
	float*preVs=(float*)malloc(numlayer*sizeof(float));
	float*nextVs=(float*)malloc(numlayer*sizeof(float));
	float*prebestVs=(float*)malloc(numlayer*sizeof(float));
	float*bestVs=(float*)malloc(numlayer*sizeof(float));
	float*pretp=(float*)malloc(numreceiver*sizeof(float));
	float*nexttp=(float*)malloc(numreceiver*sizeof(float));
	float*besttp=(float*)malloc(numreceiver*sizeof(float));
	float*prets=(float*)malloc(numreceiver*sizeof(float));
	float*nextts=(float*)malloc(numreceiver*sizeof(float));
	float*bestts=(float*)malloc(numreceiver*sizeof(float));
	float acceptpoints=0.0;
	float*ob = (float*)malloc(NUM * sizeof(float));
	float*tem = (float*)malloc(NUM * sizeof(float));
	float*bestvp = (float*)malloc(numlayer *NUM * sizeof(float));
	float*bestvs = (float*)malloc(numlayer *NUM * sizeof(float));
	time_t t;
	srand((unsigned)time(&t));
	for(i=0;i<numlayer;i++)
	{
		preVp[i]=Vop[i];
		preVs[i]=Vos[i];
		prebestVp[i]=bestVp[i]=preVp[i];
		prebestVs[i]=bestVs[i]=preVs[i];
	}


//#pragma omp parallel num_threads(NUM)//并行区内//
//	{
//		/*for (int a = 0; a < NUM; a++)
		//{*/
			/*int n = omp_get_thread_num();*/
#pragma omp parallel num_threads(NUM) private(i,k) firstprivate(preVp,preVs,nextVp,nextVs,prebestVp,prebestVs,bestVp,bestVs,temperature,acceptpoints) shared(bestvp,bestvs,ob,tem)
	{
		int n = omp_get_thread_num();
		do//annealing//
		{
			temperature *= decayscale;
			acceptpoints = 0.0;
			for (k = 0; k < markovlength; k++)//Monte Carlo sampling//
			{
				for (i = 0; i < numlayer; i++)
				{
					int Vpm = (Vpmax[i] - Vpmin[i]) / NUM;
					int Vsm = (Vsmax[i] - Vsmin[i]) / NUM;
					do
					{
						nextVp[i] = preVp[i] + stepfactor*Vpm*(rnd() - 0.5);
						nextVs[i] = preVs[i] + stepfactor*Vsm*(rnd() - 0.5);
					} while (!(nextVp[i] >= Vpmax[i] - Vpm*(n + 1) && nextVp[i] <= Vpmax[i] - Vpm*n&& nextVs[i] >= Vsmax[i] - Vsm*(n + 1) && nextVs[i] <= Vsmax[i] - Vsm*n));//constraint//
				}
				raytrace(HFt, HFb, HMt, HMb, nextVp, nextVs, Xr, Yr, Hr, numlayer, numreceiver, Hsource, Xsource, Ysource, leightmodel, nexttp, nextts);
				raytrace(HFt, HFb, HMt, HMb, bestVp, bestVs, Xr, Yr, Hr, numlayer, numreceiver, Hsource, Xsource, Ysource, leightmodel, besttp, bestts);
				if (objectfunction(besttp, bestts, top, tos, numbest, numreceiver, weight) > objectfunction(nexttp, nextts, top, tos, numbest, numreceiver, weight))
				{
					for (i = 0; i < numlayer; i++)
					{
						prebestVp[i] = bestVp[i];
						prebestVs[i] = bestVs[i];
						bestVp[i] = nextVp[i];
						bestVs[i] = nextVs[i];
					}
				}
				raytrace(HFt, HFb, HMt, HMb, preVp, preVs, Xr, Yr, Hr, numlayer, numreceiver, Hsource, Xsource, Ysource, leightmodel, pretp, prets);
				if (objectfunction(pretp, prets, top, tos, numbest, numreceiver, weight) - objectfunction(nexttp, nextts, top, tos, numbest, numreceiver, weight) > 0)
				{
					preVp[i] = nextVp[i];
					preVs[i] = nextVs[i];
					acceptpoints++;
				}
				else
				{
					float change = -1 * (objectfunction(nexttp, nextts, top, tos, numbest, numreceiver, weight) - objectfunction(pretp, prets, top, tos, numbest, numreceiver, weight)) / temperature;
					if (exp(change) > rnd())//Metropolis rule//
					{
						preVp[i] = nextVp[i];
						preVs[i] = nextVs[i];
						acceptpoints++;
					}
				}
			}
		} while (temperature > tolerance);//convergence condition//
		ob[n] = objectfunction(besttp, bestts, top, tos, numbest, numreceiver, weight);
		tem[n] = temperature;
		for (int i = 0; i < numlayer; i++)
		{
			bestvp[n*numlayer + i] = bestVp[i];
			bestvs[n*numlayer + i] = bestVs[i];
		}
	}


			/*ob[n] = objectfunction(besttp, bestts, top, tos, numbest, numreceiver, weight);
			tem[n] = temperature;
			printf("thread:%d,   temperature:%f", n, tem[n]);
			for (int i = 0; i < numlayer; i++)
			{
				bestvp[n*numlayer + i] = bestVp[i];
				bestvs[n*numlayer + i] = bestVs[i];
			}*/
		
	//}

	float obmin = ob[0];
	int bestthreadnum = 0;
	for (int i = 1; i <NUM; i++)
	{
		if (ob[i] <= obmin)
		{
			obmin = ob[i];
			bestthreadnum = i;
		}
	}
	printf("The Min value is:%f\n",obmin);
	printf("The temperature is:%f\n",tem[bestthreadnum]);
	for(i=0;i<numlayer;i++)
	{
		printf("The Min points is:%f,%f\n",bestvp[bestthreadnum*numlayer+i],bestvs[bestthreadnum*numlayer+i]);
	}
	free(preVp);
	free(nextVp);
	free(prebestVp);
	free(bestVp);
	free(preVs);
	free(nextVs);
	free(prebestVs);
	free(bestVs);
	free(pretp);
	free(nexttp);
	free(besttp);
	free(prets);
	free(nextts);
	free(bestts);
	free(ob);
	free(tem);
	free(bestvp);
	free(bestvs);
	return 0;
｝

图上是我的程序，不知为什么只要改动num增加线程后，效率大大降低，麻烦各位大神帮忙指点一下....